Transarterial Radioembolization (TARE) with Yttrium-90 (Y90) microspheres is a well-tolerated liver-directed therapy for patients with inoperable hepatocellular carcinoma (HCC). Y90 TARE uses pretreatment and post-treatment single photon emission computed tomography (SPECT)/CT for assessment of microsphere biodistribution within tumor. Patients who develop disease progression (PD) after lobar TARE have poor overall survival (OS). Conventional radiography can require several months follow-up to assess tumor response per modified RECIST (mRECIST), resulting in treatment delays for patients with PD. Predictive models capable of identifying patients at high risk for PD could prompt close surveillance and rapid initiation of salvage therapies, enhancing disease control (DC). Predictive models in various cancers have incorporated radiomics, an analytic technique that extracts digital patterns from medical imaging. We hypothesized that radiomics of immediate post-treatment SPECT/CT can predict objective response (OR) to Y90 TARE. A total of 38 lobar TARE treatments were assessed retrospectively. For all treatments, the prescribed dose was 120 Gy. SPECT/CT obtained immediately after TARE underwent radiomics analysis. A total of 75 features related to gray-level (GL) co-occurrence matrices (COM), dependency matrices (DM), run length matrices (RLM), zone size matrices (ZSM), and neighborhood difference matrices (NDM) were examined and balanced between cohorts with and without OR by aid of Gaussian noise up-sampling. Top features were chosen for a weighted k-nearest neighbors (KNN) classifier based on rank as determined by the RELIEF-F algorithm. Performance of the developed classifier was evaluated by receiver operating characteristic (ROC) curve analysis. Most patients (76%) were Child-Pugh A cirrhotic, while Barcelona Clinic Liver Cancer (BCLC) stage was evenly distributed A to C. On radiographic review, 22 treatments (58%) achieved OR. The selected top features consisted of two from CT (GLNDM-based coarseness; GLDM-based small dependence low gray level emphasis) and two from SPECT (GLNDM-based coarseness; GLZSM-based zone entropy). The weighted KNN classifier built using the selected features demonstrated a relatively strong power for predicting OR, with a ROC area under curve (AUC) of 0.83. In lobar TARE Y90 of inoperable HCC, a predictive model using texture features extracted from day of treatment SPECT/CT distinguished responders from non-responders with high accuracy. Limitations of this study include its retrospective nature and the absence of toxicity analysis. These findings suggest that predictive modeling incorporating SPECT/CT radiomics could enhance the therapeutic ratio for vulnerable HCC patients and merits further investigation in prospective clinical trials of Y90 TARE.